On Thu, 9 May 2019, Riccardo (Jack) Lucchetti wrote: > On Thu, 9 May 2019, Sven Schreiber wrote: > >> Am 09.05.2019 um 20:57 schrieb Allin Cottrell: >> >>> I agree with Sven that some clean-up and enhancement is needed, but I >>> suggest we propose that till after the 2019b release. I think what >>> might be best, when we get to it, is >>> >>> * Revert the "vif" command to just producing the VIFs (which are not >>> very sophisticated but are easy to interpret). And support a larger >>> set of estimators to the extent possible (but this would not include >>> nls, mle or gmm). The VIF vector to be available as $result. >>> >>> * Add a built-in bkw function which uses what's now in the C code to >>> replace Lee's package. Any covariance matrix accepted, so nls, mle and >>> gmm should be fine. >>> >>> * Rejig the GUI "Collinearity" menu item so that it produces what's >>> appropriate for the model in the window -- possibly with a choice of >>> VIFs, BKW or both, if applicable. >>> >>> * I think the doc should have something substantive (besides the >>> reference to Lee's paper, which is fine in itself) by way of >>> explanation of the BKW matrix, which is IMO by no means transparent. >> >> Sounds good to me! (Also the postponement.) > > I like this too.
Great, seems like we're on the same page. Adding a built-in bkw function should be quick work, the other points may take a little longer. Allin
